Prediction of milled rice grades using Fourier transform near-infrared spectroscopy and artificial neural networks

This paper describes a method to rapidly and objectively predict the grades of milled rice according to the surface lipid content (SLC), which was determined by using near-infrared (NIR) spectroscopy. Sixty-six rice varieties were milled to different degrees. Then each sample was graded by a three-member panel. After the NIR spectra for each sample were collected over the wavenumber range of 11,000-4000 cm−1, the SLC of each sample was measured according to the official method. The calibration equations relating the Fourier Transform Near-infrared (FT-NIR) spectra to the measured SLC were developed based on the partial least square (PLS) regression. The best model gave the root mean square error of the prediction (RMSEP) of 0.0248% and the determination coefficients of 0.9905. If the relationships between the grades and the SLC predicted by the developed NIR model were described with the linear and the logarithmic regression equations, the correct prediction percents (CCP) were 75.76% and 83.33%, respectively. When the back propagation artificial neural network (BP-ANN) model was developed to estimate the grades according to SLC, the resultant CCP was 95.45%, indicating that the milled rice grades could be predicted by the proposed BP-ANN model with satisfactory accuracy.